Unparalleled 5.5GHz CPU Monster

New IBM Telum IIwith 43 billion transistors in 600 mm², a true mega-processor with a total of eight highly specialized processor cores and a clock frequency of 5.5 GHz, and the IBM Spyre Accelerator According to the Armonk, New York-based company, this will open up entirely new possibilities for enterprise-level AI in the future, including large language models (“LLMs”), generative AI, and scalable I/O subsystems. It also aims to reduce power consumption and space requirements in data centers.

Mainframe chip with 360 MiByte L2 cache

With the IBM Telum II, which, like its predecessor, the IBM Telum, is a mainframe processor, or so-called “mainframe chip,” the developers clearly focused their primary focus on the highest possible capacity of fast cache memory, so-called “cache memory.” The IBM Telum II uses more than 40 percent of the chip’s massive 600 mm² (“die size”) for a total of ten chips, each containing 36 MiBytes of L2 cache, resulting in a total capacity of an astonishing 360 MiBytes of L2 cache.

Source: IBM



Overall, the IBM Telum II has exactly 40 percent more L2 cache than its predecessor, the IBM Telum, which was released in 2021.

2.88GB virtual L4 cache

But that's not enough. The IBM Telum II also features 360 MiBytes of virtual L3 cache and 2.88 GiBytes of virtual L4 cache, which can be accessed like physical cache via software specifically designed for the processor. Additionally, the processor's eight cores have been improved over their predecessors and a specially developed DPU has been integrated to speed up I/O tasks. Meanwhile, each Telum II processor can offer a 24 TOPS AI accelerator, resulting in a total computing power of 768 TOPS per system.

The high AI computing power comes from a total of 32 IBM Telum II cards, which can be installed per system and can connect a total of 192 PCIe expansion cards with the help of 12 I/O expansion cards (“expanders”), each with 16 PCIe slots. This is exactly why a specially developed DPU (“data processing unit”) is used, which can easily handle this “scalability”.

IBM Telum II and Z Series are unmatched

IBM has a very unique position thanks to the Z series and the mainframe chips embedded in it. According to the American company, 70 percent of all global financial transactions are processed via IBM Mainframe, and almost all credit card companies in the world rely on this system. This is guaranteed by a reliability of 99.999999 percent, which guarantees reliability.

Source: IBM



In addition to the IBM Telum II, the IBM Spyre Accelerator is also a separate AI accelerator, which is specified as a 75W PCIe add-on card with a 300mm² die size for a total of 300 TOPS AI workflows. Both the IBM Telum II and the IBM Spyre Accelerator, which can be completed on demand, are manufactured at Samsung on existing 5nm nodes. The IBM Telum II uses 5HPP, while the IBM Spyre uses 5LPE.

Source: IBM



IBM has dated the availability of both products for enterprise businesses to next year as part of Hot Chips 2024. As is usually the case with such products, pricing has not been officially announced.